Turn an AI agent's long-running task plan into a live visual Kanban board.
Give an agent one unambiguous next task to resume after a context reset.
Track dependencies between tasks so a checklist can't hide what's actually blocked.
Run multiple project plans as boards under one shared local hub.
| othmanadi/plandeck | 1utkarsh1/mcp-stdio-guard | oil-oil/video-publisher-skill | |
|---|---|---|---|
| Stars | 61 | 61 | 61 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 2/5 | 2/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Zero dependencies, just run the CLI against a plan.yaml file to start a board.
Plandeck turns a plan written for an AI coding agent into a visual Kanban board instead of a plain markdown checklist. The idea is that agents working on long tasks often lose track of what to do next, especially after a context reset or a manual clear, and a flat checklist cannot express that one task depends on another being finished first. Plandeck fixes this by treating the plan as a graph: each card can depend on other cards, a ready to work queue is computed automatically instead of guessed, and there is always exactly one clear next move. The board has six lanes: Backlog, Ready, In Progress, Review, Done, and Blocked. A card only becomes Blocked if its dependency is unmet or it is explicitly marked blocked, and cards do not move to Ready by hand, they get promoted automatically once their dependencies are satisfied. The longest chain of dependent cards is highlighted as the critical path so it is obvious which work is most urgent. Everything is stored in a plain YAML file on disk, so the plan survives a context reset. The computation behind the board, working out what is ready, what the critical path is, and what to do next, uses no AI or language model at all. It is built from plain, deterministic functions over the YAML data, which makes it fully testable and always gives the same answer for the same input. Running a board starts a small local web server that also acts as a hub: additional boards you start register with that same hub, so one browser tab can show every active plan at once. After a context reset, an agent can read a small file called NEXT.md, which is kept separate from the main plan and only rewritten when something actually changes, to instantly know what to work on. The project builds on an earlier project by the same author called planning-with-files and has no external dependencies. It is released under the MIT license.
Turns an AI agent's task plan into a live Kanban board with automatic dependency tracking and a critical path.
Mainly JavaScript. The stack also includes JavaScript, Node.js, YAML.
Use, modify, and distribute freely, including for commercial purposes, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.